Gemini 3.6 Flash places in the top three on 4 benchmarks.
Google DeepMind Jul 2026 Closed weights Mid-size
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LMArena (Elo) | 1,483 | #22 of 335 | Official leaderboard | |
| SEA-HELM | 76.4 | #5 of 71 | Official leaderboard | |
| OpenAI MRCR v2 | 73.9% | #4 of 47 | Official leaderboard | |
| Antidote | 1,083 | #4 of 36 | Official leaderboard | |
| Hemingway-bench | 1,075 | #5 of 42 | Official leaderboard | |
| DesignArena ASCII Art | 1,276 | #10 of 77 | Official leaderboard | |
| MRCR v2 | 91.8% | #4 of 30 | Lab model card | |
| SimpleQA Verified | 66.2% | #15 of 94 | Epoch AI run | |
| FACTS Parametric | 73.3% | #10 of 45 | Official leaderboard | |
| Global MMLU Lite | 94.3% | #10 of 43 | Official leaderboard | |
| StudyArena Text | 1,255 | #22 of 85 | Official leaderboard | |
| FACTS Multimodal | 53.7% | #6 of 23 | Official leaderboard | |
| Extended NYT Connections | 89% | #27 of 103 | Official leaderboard | |
| FACTS Parametric V2 | 73.9% | #8 of 27 | Official leaderboard | |
| Creative Writing v3 | 72.5 | #40 of 120 | Official leaderboard | |
| FACTS Benchmark Suite | 67.5% | #10 of 23 | Official leaderboard | |
| BullshitBench V2 | 39% | #65 of 129 | Official leaderboard | |
| FACTS Grounding v2 | 66.6% | #13 of 25 | Official leaderboard | |
| Document Arena | 1,454 | #25 of 39 | Official leaderboard | |
| BullshitBench | 18.2% | #75 of 107 | Official leaderboard | |
| AA-LCR Extended | 46.3% | #37 of 49 | Official leaderboard | |
| HLE Extended | 13.3% | #24 of 31 | Official leaderboard | |
| FACTS Search V2 | 68.2% | #18 of 21 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 94.1% | #13 of 229 | Epoch AI run | |
| DTBench | 95.5% | #22 of 180 | Official leaderboard | |
| ForecastBench | 0.12 | #12 of 82 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 94.2% | #54 of 212 | Epoch AI run | |
| MathArena (Apex, IMO, USAMO) | 26% | #12 of 47 | Official leaderboard | |
| ARC-AGI | 91.2% | #30 of 95 | Official leaderboard | |
| ProofBench | 36% | #21 of 59 | Official leaderboard | |
| MathArena Apex Shortlist | 71.3% | #14 of 38 | Official leaderboard | |
| AIME 2026 | 96.7% | #12 of 31 | Official leaderboard | |
| ARC-AGI-2 | 60.4% | #38 of 98 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 59% | #43 of 86 | Official leaderboard | |
| HMMT 2026 | 89.4% | #16 of 31 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 21.9% | #46 of 68 | Official leaderboard | |
| xbench-ScienceQA | 65.6% | #27 of 29 | Official leaderboard | |
| BioMysteryBench | 58.5% | #29 of 30 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DesignArena Website | 1,293 | #21 of 151 | Official leaderboard | |
| DesignArena Data Visualization | 1,288 | #17 of 118 | Official leaderboard | |
| DesignArena UI Component | 1,291 | #27 of 139 | Official leaderboard | |
| DesignArena Code | 1,280 | #31 of 143 | Official leaderboard | |
| ALE-Bench | 1,825 | #32 of 126 | Official leaderboard | |
| DesignArena 3D Design | 1,253 | #37 of 136 | Official leaderboard | |
| MLE-bench | 63.9% | #3 of 10 | Lab model card | |
| AndroidBench | 75.6% | #10 of 32 | Official leaderboard | |
| WebDev Arena | 1,537 | #33 of 105 | arena.ai, via Epoch AI | |
| DesignArena Game Dev | 1,252 | #47 of 143 | Official leaderboard | |
| WeirdML | 56.1% | #38 of 112 | Official leaderboard | |
| DesignArena Video to Website | 1,200 | #7 of 19 | Official leaderboard | |
| DesignArena Android (Kotlin) | 1,193 | #17 of 46 | Official leaderboard | |
| SWE-Bench Pro | 58.7% | #31 of 81 | Lab model card | |
| Code Arena | 1,538 | #49 of 128 | Official leaderboard | |
| ProgramBench | 0.50% | #10 of 25 | Official leaderboard | |
| DesignArena React Native | 1,192 | #24 of 56 | Official leaderboard | |
| Vibe Code Bench | 64% | #57 of 120 | Official leaderboard | |
| Image-to-WebDev Arena | 1,545 | #27 of 55 | Official leaderboard | |
| DesignArena Agentic Overall | 1,169 | #37 of 74 | Official leaderboard | |
| Code Migration | 30.9% | #41 of 78 | Official leaderboard | |
| DesignArena Full Stack | 1,159 | #26 of 49 | Official leaderboard | |
| APEX-SWE | 39.7% | #36 of 65 | Official leaderboard | |
| ProgramBench tests passed | 55.7% | #14 of 25 | Official leaderboard | |
| DesignArena Image to HTML | 1,203 | #29 of 51 | Official leaderboard | |
| DesignArena Frontend Web App | 1,180 | #31 of 53 | Official leaderboard | |
| DesignArena Image to Frontend | 1,167 | #19 of 31 | Official leaderboard | |
| DesignArena Agentic Game Dev | 1,126 | #17 of 27 | Official leaderboard | |
| ProgramBench almost solved | 4.00% | #16 of 25 | Official leaderboard | |
| SWE-bench Verified Extended | 33.3% | #33 of 44 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 4.70% | #29 of 38 | Official leaderboard | |
| DeepSWE | 46.7% | #23 of 28 | Official leaderboard | |
| IOI v2 (Vals) | 35.1% | #39 of 45 | Official leaderboard | |
| MirrorCode | 23.3% | #35 of 39 | Epoch AI run |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GameHorizon | 75.3 | #4 of 37 | Official leaderboard | |
| OSWorld-Verified | 83% | #6 of 41 | Lab model card | |
| LLM Chess | 1,094 | #20 of 123 | Official leaderboard | |
| Game Arena Coin Game | 440 | #7 of 39 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 40.1% | #10 of 44 | Official leaderboard | |
| Game Arena Chess | 1,224 | #11 of 42 | Official leaderboard | |
| Game Arena Dark Hex | 341 | #11 of 41 | Official leaderboard | |
| RuneBench | 5.58 | #22 of 76 | Official leaderboard | |
| Game Arena Four in a Row | 504 | #13 of 43 | Official leaderboard | |
| Game Arena Reversi | 710 | #13 of 39 | Official leaderboard | |
| Chess Puzzles | 42% | #36 of 93 | Epoch AI run | |
| Game Arena Poker | 2.79 | #12 of 31 | Official leaderboard | |
| Mystery Game Puzzles | 30% | #33 of 84 | Epoch AI run | |
| Long-Horizon-Terminal-Bench (LHTB) | 39% | #10 of 25 | Official leaderboard | |
| Cresta-Bench | 50.2% | #6 of 14 | Official leaderboard | |
| Game Arena Word Association | 288 | #18 of 40 | Official leaderboard | |
| Kaggle Game Arena | 327 | #19 of 42 | Official leaderboard | |
| Game Arena Word Art | 621 | #18 of 39 | Official leaderboard | |
| Game Arena Checkers | 954 | #20 of 41 | Official leaderboard | |
| AutomationBench | 17.1% | #21 of 42 | Official leaderboard | |
| DuelLab GameBench 2 | 63.1 | #30 of 57 | Official leaderboard | |
| BabaIsBench | 88% | #17 of 32 | Official leaderboard | |
| Game Arena Clobber | 290 | #21 of 39 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 269 | #21 of 39 | Official leaderboard | |
| Terminal-Bench 2.1 | 78% | #31 of 56 | Lab model card | |
| Game Arena Nine Men's Morris | 595 | #22 of 39 | Official leaderboard | |
| CLASHarness | 20.2% | #8 of 14 | Official leaderboard | |
| Game Arena Bargaining | 173 | #23 of 39 | Official leaderboard | |
| Game Arena Lines of Action | 350 | #23 of 39 | Official leaderboard | |
| Game Arena Go | 712 | #25 of 39 | Official leaderboard | |
| Game Arena Dots and Boxes | 789 | #27 of 40 | Official leaderboard | |
| BrowseComp Extended | 13.8% | #34 of 42 | Official leaderboard | |
| Bench on the Clocktower | 46.7% | #9 of 10 | Official leaderboard | |
| GAIA | 1.66% | #61 of 64 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LegalBench | 86.7% | #12 of 138 | Official leaderboard | |
| MedCode | 53.1% | #11 of 96 | Official leaderboard | |
| ProfBench | 57.4% | #9 of 72 | Official leaderboard | |
| TaxEval v2 | 74.9% | #28 of 129 | Official leaderboard | |
| Vals Finance Agent v2 | 56.3% | #20 of 80 | Official leaderboard | |
| MortgageTax | 67.6% | #22 of 87 | Official leaderboard | |
| Excel Modeling Benchmark | 65.4% | #25 of 76 | Official leaderboard | |
| APEX-1 | 66% | #24 of 71 | Official leaderboard | |
| CorpFin v2 | 63.3% | #46 of 119 | Official leaderboard | |
| Vals Index v2 | 55.4% | #28 of 65 | Official leaderboard | |
| GDPval Extended | 0.90% | #23 of 43 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 3.33% | #43 of 80 | Official leaderboard | |
| PolicyBench | 82.6% | #28 of 50 | Official leaderboard | |
| MedXpertQA MM Extended | 54.2% | #23 of 41 | Official leaderboard | |
| DesignArena PDF | 1,178 | #21 of 37 | Official leaderboard | |
| MedScribe | 79.7% | #57 of 97 | Official leaderboard | |
| APEX-Agents 1.1 | 63.4% | #39 of 65 | Official leaderboard | |
| GDPval-AA v2 | 1,421 | #17 of 27 | Lab model card | |
| YC-Bench | $968,123 | #38 of 59 | Official leaderboard | |
| FrontierFinance | 46.3% | #11 of 17 | Official leaderboard | |
| APEX-Agents | 46.9% | #40 of 61 | Collected by Epoch AI, source not named | |
| DesignArena Spreadsheet | 1,093 | #24 of 35 | Official leaderboard | |
| EnterpriseBench: CoreCraft | 39.5% | #35 of 51 | Official leaderboard | |
| Tax Agent Bench | 18% | #59 of 84 | Official leaderboard | |
| Legal Research Bench | 25% | #67 of 93 | Official leaderboard | |
| TaxCalcBench v2 | 22% | #24 of 32 | Official leaderboard | |
| DesignArena HTML Slides | 1,044 | #21 of 28 | Official leaderboard | |
| CommerceAgentBench | 46.7% | #10 of 13 | Official leaderboard | |
| DesignArena Python-PPTX Slides | 1,092 | #28 of 36 | Official leaderboard | |
| EEBench | 37.4% | #24 of 30 | Official leaderboard | |
| Public Benefits Bench | 56.6% | #41 of 50 | Official leaderboard | |
| APEX-Accounting | 52.5% | #41 of 49 | Official leaderboard | |
| OfficeQA Pro V2 | 12.2% | #7 of 7 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CharXiv | 89.4% | #9 of 75 | Lab model card | |
| Vision Arena | 1,280 | #23 of 136 | Official leaderboard | |
| 1H-VideoQA | 82.2% | #4 of 20 | Official leaderboard | |
| RoTA-Bench | 10% | #2 of 8 | Official leaderboard | |
| Math Kangaroo 2025 (MathArena) | 87.8% | #6 of 22 | Official leaderboard | |
| Chartography | 34% | #12 of 43 | Official leaderboard | |
| MMAU-Pro | 69.6% | #3 of 10 | Official leaderboard | |
| ParseBench | 69.1% | #14 of 44 | Official leaderboard | |
| ZeroBench | 18% | #25 of 76 | Official leaderboard | |
| MineBench | 1,603 | #31 of 75 | Official leaderboard | |
| VoxelBench | 1,785 | #24 of 56 | Official leaderboard | |
| vCX Benchmark | 87% | #20 of 42 | Official leaderboard | |
| Humanity's Sixth Sense | 31.9% | #17 of 34 | Official leaderboard | |
| Blueprint-Bench 2 | 31.2% | #20 of 38 | Official leaderboard | |
| Progress classification (Google DeepMind) | 43.9% | #3 of 5 | Official leaderboard | |
| CharXiv Extended | 78.4% | #25 of 38 | Official leaderboard | |
| MMMU-Pro Extended | 29% | #28 of 38 | Official leaderboard | |
| DesignArena Code to Video | 1,106 | #23 of 29 | Official leaderboard | |
| Furniture Assembly Benchmark | 23.3% | #31 of 35 | Official leaderboard | |
| Instrument reading (Google DeepMind) | 52% | #5 of 5 | Official leaderboard | |
| APM-Bench | 60.2% | #1 of 1 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 101.0 | #23 of 185 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 65.8% | #10 of 71 | Official leaderboard | |
| Human Pathogen Capabilities Test | 64% | #15 of 86 | Official leaderboard | |
| BioTIER Refuse | 81.1% | #15 of 69 | Official leaderboard | |
| BioTIER Permit | 99.8% | #16 of 69 | Official leaderboard | |
| Molecular Biology Capabilities Test | 54.3% | #23 of 90 | Official leaderboard | |
| World-Class Bio | 52.2% | #24 of 74 | Official leaderboard | |
| Big Sleep Evaluation (Google) | 42% | #2 of 3 | Official leaderboard | |
| DeepsecBench | 11.5 | #23 of 29 | Official leaderboard | |
| CyberBench (Vals) | 48.5% | #21 of 26 | Official leaderboard | |
| SAFE-Teen Health advice | 68% | #14 of 17 | Official leaderboard | |
| SAFE-Teen Urgent safety | 52% | #14 of 17 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 45.4% | #38 of 44 | Official leaderboard | |
| dfbench | 20.9% | #18 of 18 | Official leaderboard | |
| SAFE-Teen Relationship boundaries | 70% | #17 of 17 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Conceptual Reasoning Index | 59.5 | #36 of 151 | Official leaderboard | |
| LMCA | 44.9% | #43 of 147 | Official leaderboard | |
| SAGE (Vals) | 48.7% | #25 of 81 | Official leaderboard | |
| ACCoRD | 70.0 | #55 of 152 | Official leaderboard | |
| Vals Multimodal Index | 65.1% | #13 of 33 | Official leaderboard | |
| StudentBench | 12.7 | #10 of 11 | Official leaderboard |
The Self-Reported Index fits one score to the 6 capability results Google DeepMind reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 152.1 to 160.8. How we compute it.
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
Google DeepMind's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 5 of them aren't in the launch post.